Click-Gaussian is an interactive segmentation method for 3D Gaussians that enables fast and accurate segmentation of any 3D scene. The method leverages two-level granularity feature fields derived from 2D segmentation masks to facilitate segmentation without time-consuming post-processing. It addresses the challenge of inconsistent 2D segmentation masks across different views by introducing Global Feature-guided Learning (GFL), which aggregates global feature candidates across the scene to ensure consistent and distinguishable feature learning. Click-Gaussian achieves a processing speed of 10 ms per click, significantly faster than previous methods, while also improving segmentation accuracy. The method is trained on real-world scenes and can be used for various applications including object removal, resizing, repositioning, duplication, and text-based editing. The approach is evaluated on multiple datasets, demonstrating its effectiveness in both coarse and fine-level segmentation. The method's ability to handle complex scenes and provide precise segmentation makes it a promising solution for real-time 3D scene manipulation.Click-Gaussian is an interactive segmentation method for 3D Gaussians that enables fast and accurate segmentation of any 3D scene. The method leverages two-level granularity feature fields derived from 2D segmentation masks to facilitate segmentation without time-consuming post-processing. It addresses the challenge of inconsistent 2D segmentation masks across different views by introducing Global Feature-guided Learning (GFL), which aggregates global feature candidates across the scene to ensure consistent and distinguishable feature learning. Click-Gaussian achieves a processing speed of 10 ms per click, significantly faster than previous methods, while also improving segmentation accuracy. The method is trained on real-world scenes and can be used for various applications including object removal, resizing, repositioning, duplication, and text-based editing. The approach is evaluated on multiple datasets, demonstrating its effectiveness in both coarse and fine-level segmentation. The method's ability to handle complex scenes and provide precise segmentation makes it a promising solution for real-time 3D scene manipulation.